Engineering Operating System

Providing a lean single-family office with an operating system to scale and cut manual work by 64%

64%: manual workload removed
34 / 34: processes documented & digitised
six-fig.: annual licence eliminated

Hub-and-spoke: Portfolio Operating System at centre, connected to six control centres — Command Centre, Document Intelligence, and Reporting Pipeline on the left; Process Library, Relationship Graph, and Infrastructure on the right.

Case file

Industry: Single-family office (SFO)
Region: Germany
Service: Engineering Operating System / Business Case Development
Team: Wayne ArendseDelivery & Transformation LeadYurii KotulaCEO & Engineering Leader
Systems: React (Vite)
Django 6.0/DRF
PostgreSQL 16
Neo4j
DuckDB
Redis
Dagster
n8n
WireGuard VPN
local LLM (Qwen3)
Whisper.cpp

Client context

A lean German single-family office managing 12 operating companies across three industries — real estate development, professional services, and manufacturing. The office was built for control rather than scale: the founding family concentrated decision-making, and operational capacity was deliberately constrained to stay agile.

A change in business conditions required the office to grow without hiring proportionally. The expansion plan assumed a staffing constraint; the operational model did not.

Business problem

All 12 companies ran on a patchwork of spreadsheets, email threads, ad-hoc documents, and a legacy ERP licence the office could neither extend nor replace affordably. A single-family office running 12 companies cannot afford to let its knowledge live in inboxes.

The principal spent a disproportionate share of their week reconstructing state — pulling together reports, chasing updates, re-interpreting decisions made months earlier — rather than directing capital. Manual reconciliation consumed roughly 64% of operational hours that should have been spent on investment decisions and portfolio governance.

The stakes

Business continuity — a single key-person departure would leave the portfolio ungovernable
Operational resilience — any disruption to the ERP or spreadsheet chain could halt reporting
Capacity to expand — the principal could not take on new companies under the current model
Decision quality — real-time data was unavailable; decisions were made on reconstructed state

What was broken

No single source of truth across 12 companies
Knowledge locked in people, not systems
Six-figure annual licence for ERP software that covered less than 40% of operational need
Zero process documentation — 34 critical processes existed only in people's heads
Manual reconciliation consuming 64% of operational hours
No capability to onboard new portfolio companies without proportional headcount increase

Diagnostic approach

The diagnostic ran across three levels simultaneously. The family office had an operations problem, a technology problem, and a knowledge problem — in that order.

Strategy
What is the work structured around?
Modelled the portfolio structure and mapped the governance design against the expansion plan.
- Ownership model mapped
- Portfolio governance defined
- Expansion constraints documented
Operations
How does work move through the office?
Traced every operational workflow from trigger to resolution. Nothing was assumed to be working.
- All 34 processes traced end-to-end
- Bottlenecks and key-person dependencies identified
- Handoff and escalation paths documented
Technology
What does the system need to do?
Evaluated the incumbent ERP against actual operational need, not vendor capability claims.
- ERP dependency assessed and costed
- Replacement architecture specified
- Integration surface mapped

What changed

Portfolio Command Centre
A single operational view across all 12 companies: KPIs, cash positions, milestone status, and risk flags in one place. The principal's week now starts with a dashboard, not a reconstruction exercise.
Process Library
All 34 critical processes documented, structured, and linked to the systems that execute them. New staff can be onboarded in days, not months.
Document Intelligence
Contracts, reports, and correspondence indexed and searchable. A local LLM (Qwen3) runs extraction and classification without data leaving the network perimeter. Whisper.cpp handles voice notes and meeting recordings.
Relationship Graph
A Neo4j-backed relationship map connecting entities, decisions, and obligations across the portfolio. Counterparty history, open commitments, and relationship context visible without opening an inbox.
Reporting Pipeline
Automated reporting flows built on Dagster and n8n replace the manual reconciliation cycle. Reports that previously took a day to assemble now generate in minutes.
Infrastructure
WireGuard VPN with zero external cloud dependencies for sensitive data. PostgreSQL 16 as the primary operational store, DuckDB for analytical queries, Redis for real-time state.

Decisions

Local LLM over cloud AI
Sensitive counterparty and financial data cannot leave the network perimeter. Qwen3 runs entirely on-premises. Capability is slightly lower than frontier models; data sovereignty is non-negotiable.
Graph database for relationships
Counterparty and entity relationships are too complex for a relational model at this scale. Neo4j captures the real structure; joins would have hidden it.
Replace the ERP, not extend it
The incumbent licence covered less than 40% of operational need at a six-figure annual cost. Custom engineering cost less over a five-year horizon and eliminated the vendor dependency.
React (Vite) frontend
The office needed a fast, maintainable interface team members could use without training. A standard web stack with a stable component library. No abstraction layers between the UI and the operational data.

Results

Measure	Before	After
Manual reconciliation time	64% of operational hours	~0% — fully automated
Process documentation	0 of 34	34 of 34
ERP licence cost	Six-figure annual spend	Eliminated
Time to assemble weekly portfolio report	~1 working day	<5 minutes
Key-person dependency	Critical — 3 people held all operational context	Distributed across documented systems
Capacity to add a portfolio company	Not viable without new hire	Viable within existing headcount

Strike Team

Senior engineering leaders find the real constraint in a fixed-fee engineering diagnostic, then stay to fix it. Project rescue, fractional CTO, architecture and business cases.

About the Strike Team